IS IMAGENET WORTH ONE LONG VIDEO?

Summary from the TLDR AI newsletter

By training an image encoded, unsupervised, on a single long walking video, this work shows how clever model changes can result in extremely powerful representations.